JP5498889B2 - Packet relay apparatus and congestion control method - Google Patents

Description

  The present invention relates to a technique for performing congestion control in a packet relay device.

  For example, a router or a switch is known as a relay device that relays packets and frames on a network. If such a relay device has a communication request that greatly exceeds the permissible capacity, communication may be delayed inside the relay device. Such a phenomenon is generally called “congestion”. Congestion is a phenomenon that can occur in a LAN, WAN, Internet, etc., regardless of the scale of the network.

  As a technique for suppressing such congestion, there is a technique described in Patent Document 1, for example. In this technique, it is determined whether or not the value of the FIFO length counter for monitoring the number of packets buffered in the entire FIFO memory exceeds a predetermined threshold value. When the value of the FIFO length counter exceeds the threshold value, control is performed to uniformly lower the threshold value for monitoring the number of packets individually prepared for each connection. According to such control, a notification that congestion has occurred is made to the transmission source terminal connected to the connection having a larger number of packets buffered than the lowered threshold value. When receiving the notification, the transmission source terminal can reduce the transmission amount of the packet. Therefore, packets received from the connection are reduced from the FIFO memory, and congestion can be suppressed.

  However, in the technique disclosed in Patent Document 1, since the threshold value prepared for each connection is uniformly reduced, the source terminal connected to the connection having a larger number of packets buffered than the lowered threshold value All of these are notified that congestion has occurred. As a result, all the transmission source terminals simultaneously reduce the packet transmission amount, and the throughput of the entire network may be drastically reduced. Such a problem is a problem that may occur in common to apparatuses that relay data flowing on a communication line such as a network as well as a packet.

Japanese Patent Laid-Open No. 10-173661

  In light of the above-described problems, the problem to be solved by the present invention is to provide a congestion control technique capable of suppressing a decrease in the throughput of the entire network.

  SUMMARY An advantage of some aspects of the invention is to solve at least a part of the problems described above, and the invention can be implemented as the following forms or application examples.

[Application Example 1] A packet relay apparatus, which includes a buffer memory including a plurality of queues in which packets are temporarily stored, a plurality of first queue lengths respectively indicating queue lengths of the respective queues, A storage unit storing a second queue length indicating a value corresponding to a total of the first queue lengths, a receiving unit receiving a packet, and a queue to which the packet belongs from the plurality of queues A queue discriminating unit for discriminating; a queue managing unit for temporarily storing the packet in the discriminated queue in the buffer memory and updating the first queue length and the second queue length; A congestion notification determination unit that determines whether or not to perform a congestion notification to the transmission source of the packet according to the first queue length and the second queue length corresponding to the determined queue And before Packet relay apparatus and a congestion notification transmission unit that transmits the congestion notification to the sender.

  In such a configuration, based on the first queue length of the queue to which the received packet belongs and the second queue length indicating a value corresponding to the total of the plurality of first queue lengths, In order to determine whether or not to send a congestion notification to the transmission source, it becomes possible to individually perform congestion control for each of the plurality of queues. Therefore, it is possible to suppress a decrease in throughput of the entire network due to congestion control.

Application Example 2 In the packet relay apparatus according to Application Example 1, the congestion notification determination unit has a higher probability as the first queue length is longer, and is higher as the second queue length is longer. A packet relay apparatus that obtains a congestion notification probability as a probability based on a predetermined arithmetic expression and makes the determination based on the congestion notification probability.
With such a configuration, the longer the first queue length, the higher the probability of notifying the packet transmission source of congestion, and the longer the second queue length, the higher the queue length is. The probability of performing can be increased. Therefore, the probability of performing congestion notification can be obtained according to the entire queue length (second queue length) in the buffer memory and the individual queue length (first queue length) of each queue.

[Application Example 3] In the packet relay apparatus according to Application Example 2, the congestion notification determination unit is responsive to the arithmetic expression when the second queue length is equal to or smaller than a preset first threshold value. The packet relay device, wherein the congestion notification probability is 0%, and the congestion notification probability is 100% regardless of the arithmetic expression when the second queue length is equal to or greater than a preset second threshold.
With such a configuration, it is not necessary to calculate the congestion notification probability even in a situation where congestion notification is clearly required or a situation where congestion notification is not clearly required, so that the processing can be simplified.

Application Example 4 A packet relay device according to Application Example 2 or Application Example 3, comprising a setting unit that sets the arithmetic expression or a constant included in the arithmetic expression.
With such a configuration, it becomes possible to perform flexible congestion control according to the state of the network in which the packet relay device is placed and the usage state.

[Application Example 5] The packet relay apparatus according to Application Example 1, wherein the congestion notification determination unit classifies the received packet according to the first queue length, and sets the class for each class in advance. A packet relay device that performs the determination based on the determined threshold value and the second queue length.
With such a configuration, it is possible to determine whether or not to perform congestion notification only by comparing a threshold set in advance according to the first queue length and the second queue length. Congestion control can be performed individually for each of a plurality of queues while reducing the processing load on the apparatus.

[Application Example 6] The packet relay apparatus according to Application Example 5, including a setting unit that sets a threshold value for each class.
With such a configuration, it becomes possible to perform flexible congestion control according to the state of the network in which the packet relay device is placed and the usage state.

[Application Example 7] The packet relay device according to any one of Application Example 1 to Application Example 6, wherein the storage unit sets the first queue length and the second queue length to the packet Packet relay device that stores the number of data or the amount of data.
With such a configuration, the unit of the first queue length and the second queue length can be selected according to the state of the network in which the packet relay device is placed and the usage state.

[Application Example 8] The packet relay device according to any one of Application Example 1 to Application Example 7, wherein the congestion notification determination unit further determines the degree of congestion according to the second queue length. A packet relay device that determines and the congestion notification transmission unit includes the degree of congestion in the congestion notification.
With such a configuration, it is possible to provide useful information for performing congestion control on the transmission source of the packet that has received the congestion notification.

[Application Example 9] The packet relay apparatus according to Application Example 8, wherein the degree of congestion is used to limit a transmission amount of the packet at the transmission source.
With such a configuration, the packet transmission source can easily perform congestion control using the congestion notification received from the packet relay apparatus.

[Application Example 10] The packet relay device according to any one of Application Example 1 to Application Example 9, wherein the total capacity of the buffer memory is smaller than the total queue length of the plurality of queues. Relay device.
With such a configuration, packet queuing can be efficiently performed with a small capacity buffer memory based on the statistical multiplexing effect.

  In addition to the configuration as the packet relay device described above, the present invention can also be configured as a congestion control method by the packet relay device and a computer program. The computer program may be recorded on a computer-readable recording medium. As the recording medium, for example, various media such as a flexible disk, a CD-ROM, a DVD-ROM, a magneto-optical disk, a memory card, and a hard disk can be used.

It is explanatory drawing which shows schematic structure of a packet relay apparatus. It is explanatory drawing which shows the detailed structure of a congestion notification process part. It is explanatory drawing which shows the concept of a multistage queue structure. It is a flowchart of a congestion control process. It is explanatory drawing which shows an example of a 1st congestion control table. It is explanatory drawing which shows an example of the 2nd congestion control table. It is a flowchart of a 2nd congestion determination process. It is explanatory drawing which shows the concept of the calculation method of congestion notification probability with a graph. It is explanatory drawing which shows the concept of the multistage queue structure in 2nd Example. It is explanatory drawing which shows an example of the 2nd congestion control table in 2nd Example. It is a flowchart of the 2nd congestion determination process in 2nd Example. It is explanatory drawing which shows the example with which the 1st congestion control table and the 2nd congestion control table are provided in the different packet relay apparatus. It is explanatory drawing which shows the command which performs the setting regarding operation | movement of a congestion notification determination part. It is explanatory drawing which shows the command for setting a 1st congestion control table and a 2nd congestion control table.

Hereinafter, embodiments of the present invention will be described in the following order based on examples.
A. First embodiment:
A-1. Device configuration:
A-2. Congestion control processing:
A-3. First congestion determination process:
A-4. Second congestion determination process:
B. Second embodiment:
C. Third embodiment:
D. Fourth embodiment:
E. Variations:

A. First embodiment:
A-1. Device configuration:
FIG. 1 is an explanatory diagram showing a schematic configuration of a packet relay device 100 as a first embodiment of the present invention. The packet relay apparatus 100 of this embodiment includes a buffer memory 122 that temporarily stores packets received from each terminal 10 via the network line 30. Then, the packet relay apparatus 100 performs step-by-step processing of packets stored in the buffer memory 122 by a plurality of queues including queues prepared for each packet flow and queues that collectively manage these queues. to manage. That is, the packet relay apparatus 100 of this embodiment employs a shared buffer type multi-stage queue structure. The shared buffer method is a method in which a buffer memory 122 having a capacity smaller than the total capacity of each queue is shared by all the queues based on the statistical multiplexing effect. According to such a shared buffer method, it is not necessary to prepare the buffer memory 122 having a capacity corresponding to the sum of the capacity of each queue, so that the capacity of the buffer memory can be reduced, thereby reducing the component cost and power consumption Can be reduced. Hereinafter, a queue prepared for each packet flow will be described as an “N-th stage queue”, and a queue that collectively manages these queues will be described as a “(N + 1) -th stage queue”. .

  The packet relay device 100 includes a control unit 110, a memory unit 140, an interface unit 150, and a connection terminal 160. The control unit 110 is configured as a computer including a CPU, RAM, and ROM. The CPU of the control unit 110 functions as the congestion notification determination unit 120 and the congestion notification processing unit 130 by loading a predetermined control program stored in the ROM into the RAM and executing it. The congestion notification determination unit 120 and the congestion notification processing unit 130 may be realized by an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

  The memory unit 140 includes a buffer memory 122 in which packets are stored, a first congestion control table 124 for performing congestion control for the Nth stage queue, and a first congestion control table for performing congestion control for the (N + 1) th stage queue. 2 congestion control table 126. Details of the control information recorded in these tables will be described later.

  The interface unit 150 includes a plurality of physical ports 151. Each physical port 151 is connected to the terminal 10 and other packet relay devices via the network line 30.

  The congestion notification determination unit 120 receives a packet through the interface unit 150, and determines the flow of the packet based on the header information of the packet. Then, the packet is stored in a queue (Nth queue) in the buffer memory 122 corresponding to the determined flow. At this time, the congestion notification determination unit 120 also updates the first congestion control table 124 and the second congestion control table 126. A specific update method will be described later. The congestion notification determination unit 120 also has a function of transferring a packet transmitted from each queue (N-th stage queue) in the buffer memory 122 based on a predetermined scheduling method to the transmission destination of the packet through the interface unit 150. Have

  Based on the control information recorded in the first congestion control table 124 and the second congestion control table 126, the congestion notification determination unit 120 determines whether or not to notify the transmission source of the received packet. Congestion notification refers to notifying a packet transmission source so as to perform control to suppress the occurrence of congestion. When the congestion notification determination unit 120 determines to perform the congestion notification, the congestion notification determination unit 120 transmits congestion information including the address of the transmission source of the packet and information indicating the congestion state (degree of congestion) to the congestion notification processing unit 130. The congestion notification determination unit 120 corresponds to the “queue determination unit”, “queue management unit”, and “congestion notification determination unit” of the present application.

  FIG. 2 is an explanatory diagram showing a detailed configuration of the congestion notification processing unit 130. The congestion notification processing unit 130 includes a congestion notification packet generation unit 132 and a congestion notification packet transmission unit 134. The congestion notification packet generation unit 132 generates a packet (congestion notification packet) for performing congestion notification based on the congestion information received from the congestion notification determination unit 120. This congestion notification packet includes an address of a packet transmission destination and information indicating a congestion state. The congestion notification packet transmission unit 134 transmits the congestion notification packet to the corresponding address through the interface unit 150. The transmission source of the packet that has received the congestion notification packet limits the amount of packet transmission according to the notified degree of congestion.

  The management device 20 is connected to the connection terminal 160 (FIG. 1) via a predetermined cable. The management device 20 is connected to a display device such as a display and an input device such as a keyboard and a mouse. A user who manages the packet relay apparatus 100 (hereinafter referred to as “management user”) can use the management apparatus 20 to make various settings for the packet relay apparatus 100. A specific example of the setting contents will be described later. As an interface for connecting the connection terminal 160 and the management apparatus 20, various standards can be applied. For example, a serial interface, a LAN interface, a USB interface, or the like can be applied.

  FIG. 3 is an explanatory diagram showing the concept of the multistage queue structure in the present embodiment. FIG. 3 shows an example in which 1000 queues exist in the Nth stage. In the present embodiment, each of the plurality of queues in the Nth stage individually corresponds to a packet flow. However, the queue of the Nth stage is not limited to the packet flow, but can be individually associated with the packet transmission source user and the physical port 151. Specifically, for example, a packet from a certain user A is always stored in the Nth queue # 1, and a packet from a certain user Z is always stored in the Nth queue # 1000. Is possible.

  In this embodiment, a shared buffer type multi-stage queue structure using a statistical multiplexing effect is employed. Therefore, the total number of packets stored in each of the N-th stage queues is managed by the (N + 1) -th stage queues. Specifically, the current queue length QLEN (N + 1) of the (N + 1) -th queue is the total value of the current queue lengths QLEN (N) of a plurality of N-th queues. FIG. 3 shows a congestion notification threshold CN_TH (N) for determining a congestion state in the N-th stage queue, and a maximum congestion notification threshold MAX_CN_TH (N +) for determining a congestion state in the (N + 1) -th stage queue. 1) and a minimum congestion notification threshold value MIN_CN_TH (N + 1). The usage of these threshold values will be described later.

A-2. Congestion control processing:
FIG. 4 is a flowchart of the congestion control process that the packet relay apparatus 100 repeatedly executes. This congestion control processing is executed simultaneously and in parallel with processing for the packet relay device 100 to queue and relay packets. Since processing for queuing and relaying packets is well known, description thereof is omitted in this embodiment.

  When the execution of the congestion control process is started, the congestion notification determination unit 120 receives a packet through the interface unit 104 (step S100). Then, the congestion notification determination unit 120 determines the flow by analyzing the header information of the received packet, and determines to which queue of the N-th stage queue the packet belongs (step S200).

  When the congestion notification determination unit 120 determines the queue to which the packet belongs (Nth queue), the congestion notification determination unit 120 determines the queue (Nth queue) based on the control information recorded in the first congestion control table 124 in the memory unit 140. It is determined whether or not the queue is congested (step S300). Hereinafter, this determination process is referred to as a first congestion determination process. Details of the first congestion determination process will be described later.

  When it is determined by the first congestion determination process that congestion has occurred in the N-th queue (step S400: Yes), the congestion notification determination unit 120 determines whether the congestion has been determined by the first congestion determination process. Congestion information including the degree and information indicating the address of the transmission source of the packet received in step S100 is notified to the congestion notification processing unit 130. Upon receiving this notification, the congestion notification processing unit 130 generates a congestion notification packet based on the notified congestion information (step S700), and transmits the congestion notification packet to the transmission source of the packet received in step S100 (step S700). S800).

  When it is determined by the first congestion determination process in step S300 that congestion has not occurred in the N-th queue (step S400: No), the congestion notification determination unit 120 further performs the second congestion control. Based on the control information recorded in the table 126, a process of determining whether or not congestion has occurred in the (N + 1) -th stage queue (hereinafter referred to as “second congestion determination process”) is executed (step S500). ). Details of the second congestion determination process will be described later.

  When it is determined by the second congestion determination process that congestion has occurred in the (N + 1) -th stage queue (step S600: Yes), the congestion notification determination unit 120 is determined by the second congestion determination process. The congestion notification processing unit 130 is notified of congestion information including the degree of congestion and information indicating the address of the transmission source of the packet received in step S100. Upon receiving this notification, the congestion notification processing unit 130 generates a congestion notification packet based on the notified congestion information (step S700), and transmits the congestion notification packet to the transmission source of the packet received in step S100 (step S700). S800).

  On the other hand, when it is determined by the second congestion determination processing that no congestion has occurred in the (N + 1) -th stage queue (step S600: No), the processing of steps S700 and S800 is skipped and the congestion The control process ends.

A-3. First congestion determination process:
Subsequently, specific processing contents of the first congestion determination processing described above will be described.
FIG. 5 is an explanatory diagram showing an example of the first congestion control table 124 used in the first congestion determination process. The first congestion control table 124 includes a maximum storage threshold MAX_TH (N) representing the maximum number of packets that can be stored in each queue for each N-th queue represented by an individual queue number, and whether congestion notification is performed. A congestion notification threshold CN_TH (N) for determining whether or not and a queue length QLEN (N) indicating the number of packets currently stored in the queue are recorded as control information.

  The congestion notification determination unit 120 updates the first congestion control table 124 as follows. That is, the congestion notification determination unit 120 stores the received packet in the corresponding queue in the buffer memory 122 and simultaneously increases the queue length QLEN (N) of the corresponding queue number in the first congestion control table 124 by one. When a packet is later sent from the queue in the buffer memory 122, the congestion notification determination unit 120 decreases the queue length QLEN (N) of the corresponding queue number in the first congestion control table 124 by one.

  In the first congestion determination process described above, the congestion notification determination unit 120 determines that the queue length QLEN (N) of the queue corresponding to the received packet is less than the congestion notification threshold CN_TH (N) set for the queue. It is determined that no congestion has occurred in the queue. On the other hand, if the queue length QLEN (N) of the queue corresponding to the received packet is equal to or greater than the congestion notification threshold CN_TH (N), the congestion notification determination unit 120 determines that congestion has occurred in the queue. .

  For example, in the example shown in FIG. 5, in queue number # 4, the queue length QLEN (N) is “90”, which exceeds the congestion notification threshold CN_TH (N) “80”. The notification determination unit 120 determines that congestion has occurred in the queue. On the other hand, since the queue lengths QLEN (N) of all other queues are all less than each congestion notification threshold CN_TH (N), the congestion notification determination unit 120 determines that no congestion has occurred for the other queues. To do. If the congestion notification determination unit 120 determines that congestion has occurred, the congestion notification determination unit 120 calculates the ratio of the queue length QLEN (N) to the congestion notification threshold CN_TH (N), and determines the degree of congestion according to this ratio. To do. The degree of congestion can be determined based on a predetermined table such as “light congestion” and “heavy congestion”.

A-4. Second congestion determination process:
Next, specific processing contents of the above-described second congestion determination processing will be described.
FIG. 6 is an explanatory diagram showing an example of the second congestion control table 126 used in the second congestion determination process. The second congestion control table 126 includes the (N + 1) -th queue number (one in this embodiment) and the maximum number of packets that can be stored in the (N + 1) -th queue (that is, the entire buffer memory 122). Representing the maximum storage threshold MAX_TH (N + 1), and the maximum congestion notification threshold MAX_CN_TH (N + 1) indicating the queue length to be notified of congestion to the transmission source of all packets transmitted to the queue, The minimum congestion notification threshold MIN_CN_TH (N + 1) that indicates the upper limit of the queue length for which congestion notification is not performed, and the probability that congestion notification will be performed when the average queue length QAVE described later reaches the maximum congestion notification threshold MAX_CN_TH (N + 1) Reference congestion notification probability STD_CP, a queue length QLEN (N + 1) indicating the total number of packets stored in the N-th stage queues, and a queue length QLEN (N +1) Average queue length QAVE indicating the average value is recorded as control information Is done. In FIG. 6, the maximum storage threshold MAX_TH (N + 1) is 1000, the maximum congestion notification threshold MAX_CN_TH (N + 1) is 800, the minimum congestion notification threshold MIN_CN_TH (N + 1) is 500, and the reference congestion notification probability STD_CP is 50. %, Queue length QLEN (N + 1) is 700, and average queue length QAVE is 700. In this embodiment, since the shared buffer method using the statistical multiplexing effect is adopted, the maximum storage threshold MAX_TH (N + 1) of the (N + 1) -th queue is the N-th stage shown in FIG. The value is smaller than the total value of the maximum queue storage threshold MAX_TH (N).

  The congestion notification determination unit 120 updates the second congestion control table 126 as follows. That is, the congestion notification determination unit 120 stores the received packet in the corresponding N-th stage queue in the buffer memory 122 and simultaneously increases the queue length QLEN (N + 1) in the second congestion control table 126 by one. Further, the average queue length QAVE is updated by the method described later. When a packet is transmitted later from the buffer memory 122, the congestion notification determination unit 120 decreases the queue length QLEN (N + 1) in the second congestion control table 126 by one and sets the average queue length QAVE to Update.

  FIG. 7 is a flowchart showing specific processing contents of the second congestion determination processing. When the second congestion determination process is started, the queue length QLEN (N + 1) of the second congestion control table 126 is updated with the reception of the packet. The average queue length QAVE in the congestion control table 126 is updated (step S510). The updated average queue length QAVEnew can be obtained by the following equation (1) based on the current queue length QLEN (N + 1) and the average queue length QAVEold before update.

QAVEnew = {QLEN (N + 1) * 1/2 ^ S} + {QAVEold * (1-1 / 2 ^ S)} (1)
(However, S is a weighting factor larger than 0.)

  According to the above equation, in this embodiment, the larger the weighting factor S, the smaller the influence of the current queue length QLEN (N + 1) on the updated average queue length QAVEnew. That is, if the weighting factor S is increased, even if the current queue length QLEN (N + 1) varies greatly within a short period, the variation of the average queue length QAVE can be decreased.

  Subsequently, the congestion notification determination unit 120 acquires the queue length QLEN (N) of the N-th queue to which the received packet belongs and the congestion notification threshold CN_TH (N) from the first congestion control table 124 (step S520). When the queue length QLEN (N) and the congestion notification threshold CN_TH (N) are acquired from the first congestion control table 124, the congestion notification determination unit 120 obtains the acquired queue length QLEN (N) and the congestion notification threshold CN_TH (N). Based on the control information recorded in the second congestion control table 126, the congestion notification probability CP is calculated (step S530). With this calculation process, the congestion notification probability CP takes a different value for each queue (N-th stage queue) to which the received packet belongs.

  FIG. 8 is an explanatory diagram showing a concept of a method for calculating the congestion notification probability CP by a graph. The congestion notification determination unit 120 first calculates the congestion notification probability CP by calculating the average queue length QAVE calculated in step S510 and the maximum congestion notification threshold MAX_CN_TH (N + 1) recorded in the second congestion control table 126. Depending on the minimum congestion notification threshold MIN_CN_TH (N + 1), the calculation formulas are divided into cases. Specifically, case classification is performed as in the following (A), (B), and (C). Then, according to each case, the congestion notification probability CP is calculated based on the following formulas (2) to (4). In FIG. 8, the range of values that can be taken by the congestion notification probability CP calculated in this way is indicated by a solid line and hatching.

(A) When QAVE ≧ MAX_CN_TH (N + 1):
CP = 100% (2)
(B) If QAVE ≦ MIN_CN_TH (N + 1) -β:
CP = 0% (3)
(C) When MIN_CN_TH (N + 1) -β <QAVE <MAX_CN_TH (N + 1):
CP = (STD_CP + α) (QAVE-MIN_CN_TH (N + 1) + β) / (MAX_CN_TH (N + 1) -MIN_CN_TH (N + 1) + β) (4)
(Where α is a congestion notification probability weight and β is a congestion notification threshold weight.)

  As shown in FIG. 8, the congestion notification probability CP when the average queue length QAVE matches the maximum congestion notification threshold MAX_CN_TH (N + 1) is the absolute value when the congestion notification probability weight α is a positive value. Is larger than the reference congestion notification probability STD_CP. On the other hand, the congestion notification probability CP when the average queue length QAVE matches the maximum congestion notification threshold MAX_CN_TH (N + 1), the reference value increases as the absolute value increases when the congestion notification probability weight α is a negative value. It becomes smaller than the congestion notification probability STD_CP.

  When the congestion notification threshold weight β is a positive value, the threshold value at which the congestion notification is started becomes smaller than the minimum congestion notification threshold MIN_CN_TH (N + 1) as the absolute value increases. On the other hand, when the congestion notification threshold weight β is a negative value, the threshold value at which the congestion notification is started becomes larger than the minimum congestion notification threshold MIN_CN_TH (N + 1) as the absolute value increases. In other words, in this embodiment, the relationship between the congestion notification probability CP and the average queue length QAVE that are determined that the (N + 1) -th stage queue is in a congestion state according to the congestion notification probability weight α and the congestion notification threshold weight β. Fluctuates.

  Note that the congestion notification probability weight α and the congestion notification threshold weight β are based on the queue length QLEN (N) and the maximum storage threshold MAX_TH (N) of the Nth queue corresponding to the received packet. ), (6). Note that the calculation method of the congestion notification probability weight α and the congestion notification threshold weight β is not limited to the following formulas, and various methods can be adopted.

α = αMAX * {QLEN (N) / MAX_TH (N) * 2-1} (5)
β = βMAX * {QLEN (N) / MAX_TH (N) * 2-1} (6)
(Here, αMAX and βMAX are predetermined constants.)

  According to the above formulas (5) and (6), the larger the ratio of the queue length QLEN (N) to the maximum congestion notification threshold MAX_TH (N) of the N-th queue to which the received packet belongs, the greater the congestion notification probability weight. α and congestion notification threshold weight β increase. Therefore, the greater the ratio of the queue length QLEN (N) to the maximum congestion notification threshold MAX_TH (N) of the Nth queue to which the received packet belongs, the greater the congestion in the (N + 1) th queue. The probability of determination (congestion notification probability CP) increases.

  Note that the value of αMAX in Expression (5), the value of βMAX in Expression (6), and the value of the weighting coefficient S in Expression (1) are set to arbitrary values by the management device 20 connected to the connection terminal 160. Can be possible. By doing so, the management user can freely set and change the range in which the congestion notification probability weight α and the congestion notification threshold weight β vary, and the weight coefficient S. For example, the management user can appropriately adjust the fluctuation range of the congestion notification weight α and the congestion notification threshold weight β and the weighting factor S according to the number of users connected to the packet relay device 100, the communication amount, and the like. Further, the values of αMAX and βMAX and the weighting coefficient S may be selectable by the management user from a plurality of values prepared in advance. Further, the management user can arbitrarily set each value set in the first congestion control table 124 and the second congestion control table 126 by using the management device 20.

  When the congestion notification probability CP is calculated as described above, the congestion notification determination unit 120, based on the calculated congestion notification probability CP, the (N + 1) -th queue (in other words, the entire buffer memory 122). It is determined whether or not congestion has occurred (step S540). Specifically, for example, as shown in FIG. 6, the maximum storage threshold MAX_TH (N + 1) is 1000, the maximum congestion notification threshold MAX_CN_TH (N + 1) is 800, and the minimum congestion notification threshold MIN_CN_TH (N + 1). Is 500, the reference congestion notification probability STD_CP is 50%, the queue length QLEN (N + 1) is 700, the average queue length QAVE is 700, the constant αMAX is 40%, and the constant βMAX is 400. Then, when the queue length QLEN (N) of the N-th queue to which the packet belongs is 3/4 of the maximum storage threshold MAX_TH (N), the congestion notification probability weight α is 20% and the congestion notification threshold weight β is 200. Is calculated. Since the average queue length QAVE of the (N + 1) -th stage queue is 700, the congestion notification probability CP in the (N + 1) -th stage queue of the packet is calculated as 56%. Therefore, in step S540, it is determined that congestion has occurred with a probability of 56%. In addition, for example, when the queue length QLEN (N) of the N-th queue to which the packet belongs is 1/8 of the maximum storage threshold MAX_TH (N), the congestion notification probability weight α is −30%, the congestion notification threshold The weight β is calculated as −300. Further, since the average queue length QAVE of the (N + 1) -th stage queue is 700, the congestion notification probability CP in the (N + 1) -th stage queue of the packet is 0%. For this reason, in step S540, it is determined that no congestion has occurred in the queue.

  When it is determined whether or not congestion has occurred in the (N + 1) -th queue as described above, the congestion notification determination unit 120 then determines the average queue length for the maximum congestion notification threshold MAX_CN_TH (N + 1). The ratio of QAVE (which may be queue length QLEN (N + 1)) is obtained, and the degree of congestion is determined according to this ratio in the same manner as in the first congestion determination process (step S550). Through the series of processes described above, the second congestion determination process ends, and the process returns to the congestion control process of FIG.

  According to the first embodiment described above, the congestion notification probability CP for the N-stage queue having a large number of packets stored in the buffer memory 122 among the plurality of N-stage queues increases. The congestion notification probability CP for the N-th queue with a small number of packets stored in the buffer memory 122 is reduced. That is, in this embodiment, for each Nth stage queue, the probability of performing congestion notification dynamically changes according to the queue length. Therefore, even if the number of packets in a specific queue at the N-th stage increases, the congestion notification probability CP for that queue individually increases. Congestion notification can be performed. Thereby, in the congestion control based on the queue in the subsequent stage, control reflecting the state of each queue in the previous stage is possible, and communication from the transmission source causing the congestion can be accurately restricted. As a result, no restriction is imposed on a transmission source that does not cause congestion, so that it is possible to suppress a decrease in throughput of the entire network. That is, according to the present embodiment, it is possible to suppress the occupation of the shared buffer by a specific queue, and it is possible to prevent communication from being inhibited because other queues cannot use the shared buffer. In addition, since the transmission source that has received the congestion notification can reduce the amount of communication according to the notified congestion state, it is also possible to reduce power consumption.

  In this embodiment, as shown in FIG. 4, the first congestion determination process is executed for the N-th stage queue, and the second congestion determination process is executed for the (N + 1) -th stage queue. . However, the execution of the first congestion determination process may be omitted. Even in this case, only by executing the second congestion determination process, the packet relay apparatus 100 operates as if the gradual congestion control is performed individually in all the queues. In the second congestion determination process, the queue length QLEN (N + 1) can be used for calculating the congestion notification probability CP instead of the average queue length QAVE.

B. Second embodiment:
FIG. 9 is an explanatory diagram showing the concept of the multi-stage queue structure in the second embodiment. The difference from the first embodiment described above is that the concept of “congestion class” is used in the congestion notification determination in the (N + 1) -th stage queue, and the first congestion determination process executed for the N-th stage queue is different. The processing contents and configuration are the same as in the first embodiment.

  The “congestion class” shown in FIG. 9 indicates the priority of the packet stored in the queue according to the state of the N-th queue to which the received packet belongs (specifically, the queue length QLEN (N)). Classify. In this embodiment, when a received packet is classified as a high priority, congestion notification is made difficult for the packet transmission source, and when it is classified as a low priority, the packet transmission source is congested. Control to make notifications easier. That is, in this embodiment, the ease of notification of congestion to the packet transmission source varies according to the classified congestion class.

  For example, in the case of FIG. 9, since the queue length QLEN (N) of the N-th queue # 1000 is relatively short, the congestion notification determination unit 120 determines that the packets belonging to the N-th queue # 1000 Assign to high congestion class 4. On the other hand, the queue # 1 at the Nth stage has a relatively long queue length QLEN (N), so the congestion notification determination unit 120 assigns packets belonging to the queue # 1 at the Nth stage to the congestion class 2 having a low priority. . That is, packets belonging to the N-th stage queue having a long queue length QLEN (N) are classified into a low-priority congestion class. In the second congestion determination process in the (N + 1) -th stage queue, The possibility of being determined to be in a congestion state increases. As a result, the buffer memory 122 can be prevented from being occupied by a specific queue having a long queue length QLEN (N).

  FIG. 10 is an explanatory diagram showing an example of the second congestion control table 126b in the second embodiment. In the second congestion control table 126b in the second embodiment, the maximum storage threshold MAX_TH (N + 1) representing the maximum number of packets that can be stored in the (N + 1) -th queue (that is, the entire buffer memory 122), and N A queue length QLEN (N + 1) indicating the total number of packets stored in the queue in the stage is stored. In the present embodiment, the four congestion notification thresholds TH # 1 to TH # 4 respectively set for the congestion classes 1 to 4 are stored in the second congestion control table 126b. As shown in FIG. 9, the congestion notification thresholds TH # 1 to TH # 4 are set to different values for each of the congestion classes 1 to 4, and the congestion class with higher priority is set to a larger value. Yes.

  FIG. 11 is a flowchart of the second congestion determination process in the second embodiment. When the second congestion determination process is started in the present embodiment, the congestion notification determination unit 120 first determines the congestion class (N) of the packet according to the queue length QLEN (N) of the N-th queue to which the received packet belongs. 1-4) is determined (step S510b). For the congestion class, for example, the ratio of the queue length QLEN (N) to the congestion notification threshold CN_TH (N) (or the maximum storage threshold MAX_TH (N)) set for the queue is obtained, and if the obtained ratio is up to 25% Class 1 can be determined according to the ratio of class 2 up to 50%, class 2 up to 75%, class 4 up to 75% or more.

  When the congestion class is determined, the congestion notification determination unit 120 sets the congestion notification threshold associated with the determined congestion class and the queue length QLEN (N + 1) of the (N + 1) -th stage queue to the second congestion control. Obtained from the table 126b (step S520b), and compares them (step S540b). As a result of this comparison, when the queue length QLEN (N + 1) is equal to or greater than the congestion notification threshold corresponding to the congestion class, the congestion notification determination unit 120 determines that the queue is in a congestion state and performs congestion notification. And decide. On the other hand, if the queue length QLEN (N + 1) is less than the congestion notification threshold corresponding to the congestion class, it is determined that the queue is not in a congestion state. Finally, the congestion notification determination unit 120 calculates a ratio of the queue length QLEN (N + 1) to the congestion notification threshold, and determines the degree of congestion according to this ratio (step S550b). Through the series of processes described above, the second congestion determination process in the second embodiment is completed, and the process returns to the congestion control process shown in FIG.

  In the second embodiment described above, the congestion class is determined by the queue length QLEN (N) of the N-th queue to which the packet belongs, and the congestion notification threshold and queue length QLEN (N +) set for each congestion class are concerned. Based on 1), the congestion notification determination in the (N + 1) -th queue is performed. That is, since it is possible to determine whether or not congestion notification is to be performed for each queue by simple comparison between the queue length QLEN (N + 1) and a threshold value corresponding to the congestion class, the processing load on the packet relay device 100 is reduced. However, the same effect as the first embodiment can be obtained.

  Note that the values of the congestion notification threshold values TH # 1 to TH # 4, the congestion class assignment method, and the number of congestion classes shown in FIG. 10 are not limited to the above-described example, and can be arbitrarily set. Further, the management user can arbitrarily set these by the management device 20 connected to the connection terminal 160.

C. Third embodiment:
In the first embodiment and the second embodiment described above, one packet relay device 100 includes the first congestion control table 124 and the second congestion control table 126. In contrast, in the third embodiment, the first congestion control table 124 and the second congestion control table 126 are provided in different packet relay apparatuses.

  FIG. 12 is an explanatory diagram illustrating an example in which the first congestion control table 124 and the second congestion control table 126 are provided in different packet relay apparatuses. In the present embodiment, the packet relay devices 100A and 100B manage the queue at the Nth stage, and the packet relay device 100C manages the queue at the (N + 1) th stage. The packet relay devices 100A and 100B are connected to the packet relay device 100C via the arbitration device 111. The arbitrating device 111 is a device for transferring control information transmitted from the packet relay devices 100A and 100B to the packet relay device 100C. Each packet relay device and the arbitrating device may be connected by a network or may be connected by a predetermined bus.

  The packet relay apparatuses 100A and 100B include buffer memories 122A and 122B and first congestion control tables 124A and 124B in the memory units 140A and 140B. The packet relay device 100C includes a buffer memory 122C and a second congestion control table 126C in the memory unit 140C. That is, in this embodiment, an individual buffer system configuration in which buffer memories are physically separated is employed. Note that the packet relay apparatuses 100A and 100B can manage different queues in the Nth stage.

  Hereinafter, the congestion control in this embodiment will be described. First, when the packet relay apparatuses 100A and 100B receive a packet, the congestion notification determination units 120A and 120B determine to which queue the packet belongs. The congestion notification determination units 120A and 120B perform the same process as the first congestion determination process in the first embodiment described above. When it is determined by the first congestion determination process that the queue to which the received packet belongs is in a congestion state, the congestion notification processing units 130A and 130B notify the congestion to the packet transmission source. On the other hand, when it is determined that the received packet is not in a congestion state, no congestion notification is made.

  After the first congestion determination process, when the packet stored in the N-th stage queue is sent out, the data of the packet is read by the arbitration device 111 and the queue of the packet queue The long QLEN (N) is read from the first congestion control table 124A or 124B and transmitted to the packet relay device 100C. At the same time, the queue length QLEN (N) in the first congestion control tables 124A and 124B decreases by one, and the queue length QLEN (N + 1) in the second congestion control table 126C increases by one. In the packet relay device 100C that has received the packet and the queue length QLEN (N), the same processing as the second congestion determination processing of the first embodiment described above is executed, thereby causing congestion notification based on the (N + 1) -th stage queue. It is determined whether or not to perform.

  As in the third embodiment described above, even if the buffer memory is physically divided between the N-th queue and the (N + 1) -th queue, the sharing in the first and second embodiments. Similar to the buffer method, the congestion notification determination at the (N + 1) -th stage can be realized by reflecting the state of the queue at the N-th stage.

  In this embodiment, the example in which the individual buffer method is adopted in the plurality of packet relay apparatuses has been described. However, the individual buffer method may be adopted in one packet relay device. In this case, all the components shown in FIG. 12 are provided in one packet relay device.

D. Fourth embodiment:
In each of the above-described embodiments, the management device 20 connected to the connection terminal 160 sets the settings related to the operation of the congestion notification determination unit 120 and the control information in the first congestion control table 124 and the second congestion control table 126. Is possible. Therefore, finally, a specific example of a command for performing these settings will be shown.

  FIG. 13 is an explanatory diagram illustrating commands for performing settings related to the operation of the congestion notification determination unit 120. FIG. 13 shows commands for setting the reference congestion notification probability STD_CP, the weight coefficient S, the congestion notification probability weight α, the congestion notification threshold weight β, the number of congestion classes, and the threshold value of each congestion class. When the congestion notification determination unit 120 receives these commands from the management device 20, the congestion notification determination unit 120 sets each control information according to the received commands. Further, as shown in FIG. 13, a command for displaying these control information and a command for clearing are also prepared.

  FIG. 14 is an explanatory diagram showing commands for setting the first congestion control table 124 and the second congestion control table 126. In FIG. 14, the control information shown in FIGS. 5 and 6 such as the number of queues at the Nth stage and the maximum congestion notification threshold MAX_CN_TH (N + 1) of the (N + 1) th stage queue is set. The command was shown. Upon receiving these commands from the management device 20, the congestion notification determination unit 120 sets each control information in the first congestion control table 124 and the second congestion control table 126 according to the received command. Further, as shown in FIG. 14, a command for displaying the control information and a command for clearing are also prepared.

  As described above, the settings related to the operation of the congestion notification determination unit 120 and the values in the first congestion control table 124 and the second congestion control table 126 are arbitrarily set by the management device 20 connected to the connection terminal 160. If the setting is possible, fine congestion control can be realized according to the number of terminals connected to the packet relay device 100, the traffic, the state of the network, the usage status of the packet relay device 100, and the like.

E. Variations:
As mentioned above, although several Example of this invention was described, this invention is not limited to these Examples, A various structure can be taken in the range which does not deviate from the meaning. For example, a function realized by software may be realized by hardware, and conversely, a function realized by hardware may be realized by software. In addition, the following modifications are possible.

・ Modification 1:
In each of the above-described embodiments, packets accumulated in the Nth queue and the (N + 1) th queue are managed by the number of packets. That is, each value stored in the first congestion control table 124 or the second congestion control table 126 represents the number of packets. On the other hand, the management of each queue may be performed according to the communication amount (for example, the number of bytes).

Modification 2
In each of the above-described embodiments, the configuration having the two-stage queues of the N-th stage queue and the (N + 1) -th stage queue has been described. The same can be applied. Specifically, in determining congestion notification of a packet in each stage queue, the congestion notification determination unit 120 performs congestion notification determination based on the control information of the queue before the previous stage and the control information of the queue at that stage. For example, there are a plurality of types of first congestion control tables 124 for managing the N-th stage queue, and second congestion control tables 126 for managing the (N + 1) -th stage queue are included in the respective first congestion control tables 124. On the other hand, if one is prepared, a third congestion control table for managing these second congestion control tables 126 as a queue at the (N + 2) stage can be prepared. In this case, in the (N + 1) -th stage queue, as in the first congestion control process described above, the congestion determination process is simply performed by comparing the (N + 1) -th stage queue length with a predetermined threshold value. Alternatively, as in the second congestion control process described above, a congestion determination process reflecting the state of the previous queue may be performed.

・ Modification 3:
In the first embodiment, the congestion notification determination unit 120 determines to which of the plurality of N-th queues the packet received by the packet relay device 100 belongs. However, this determination may be performed by another device different from the packet relay device 100. Moreover, it is good also as a functional block or circuit different from the congestion notification determination part 120 performing.

-Modification 4:
The formula for obtaining the congestion notification probability CP and the formula for obtaining the congestion notification probability weight α and the congestion notification threshold weight β in the first embodiment are examples, and control information in the queue at the Nth stage is used as a parameter. Other formulas may be used as long as they are included. Also, an expression for determining the congestion notification probability CP, an expression for determining the congestion notification probability weight α and the congestion notification threshold weight β, constants included in these expressions, and each congestion class in the second embodiment are set. The congestion notification threshold value may be changed by an instruction from the management user via the management device 20, or may be selected from a plurality of options adjusted in advance. This makes it possible to appropriately adjust the packet congestion notification determination method according to changes in the number of users and traffic.

Modification 5:
In the above embodiment, the management device 20 is connected to the connection terminal 160 and various settings are performed from the management device 20. On the other hand, a display device such as a display or an input device such as a keyboard may be directly connected to the packet relay device 100 and settings may be performed using these. Of course, the packet relay device 100 itself may include a display device and an input device.

DESCRIPTION OF SYMBOLS 10 ... Terminal 20 ... Management apparatus 30 ... Network line 100, 100A, 100B, 100C ... Packet relay apparatus 104 ... Interface part 110 ... Control unit 111 ... Arbitration apparatus 120, 120A, 120B, 120C ... Congestion notification determination part 122, 122A, 122B, 122C ... Buffer memory 124, 124A, 124B ... First congestion control table 126, 126C ... Second congestion control table 126b ... Second congestion control table 130, 130A, 130B, 130C ... Congestion notification processing unit 132 ... Congestion notification packet Generation unit 134 ... Congestion notification packet transmission unit 140, 140A, 140B, 140C ... Memory unit 150 ... Interface unit 151 ... Physical port 160 ... Connection terminal

Claims (10)

A packet relay device,
Buffer memory containing multiple queues where packets are temporarily stored;
A storage unit for storing a plurality of first queue lengths indicating the queue lengths of the respective queues, and a second queue length indicating a value corresponding to a total of the plurality of first queue lengths;
A receiver for receiving the packet;
A queue determination unit for determining a queue to which the packet belongs from the plurality of queues;
A queue management unit for temporarily storing the packet in the determined queue in the buffer memory and updating the first queue length and the second queue length;
A congestion notification determination unit that determines whether or not to perform congestion notification to the transmission source of the packet according to the first queue length corresponding to the determined queue and the second queue length; ,
A congestion notification transmitter for transmitting the congestion notification to the transmission source;
Equipped with a,
The congestion notification determination unit classifies the received packet according to the first queue length, and performs the determination based on a threshold set in advance for each class and the second queue length. Do,
Packet relay device. The packet relay device according to claim 1,
The congestion notification determination unit obtains a congestion notification probability that is higher as the first queue length is longer and is higher as the second queue length is longer based on a predetermined arithmetic expression. A packet relay device that performs the determination based on a notification probability. The packet relay device according to claim 2,
The congestion notification determination unit sets the congestion notification probability to 0% regardless of the arithmetic expression when the second queue length is equal to or less than a preset first threshold, and the second queue length is set in advance. A packet relay device in which the congestion notification probability is set to 100% regardless of the arithmetic expression when the value is equal to or greater than the set second threshold value. The packet relay device according to claim 2 or 3, wherein
A packet relay device comprising a setting unit that sets the arithmetic expression or a constant included in the arithmetic expression. The packet relay device according to any one of claims 1 to 4 , wherein
A packet relay apparatus comprising a setting unit for setting a threshold value for each class. The packet relay device according to any one of claims 1 to 5 , wherein
The packet relay device, wherein the storage unit stores the first queue length and the second queue length according to the number of packets or the amount of data. The packet relay device according to any one of claims 1 to 6 , wherein
The congestion notification determination unit further determines the degree of congestion according to the second queue length,
The congestion notification transmission unit is a packet relay device that includes the degree of congestion in the congestion notification. The packet relay device according to claim 7 ,
The degree of congestion is a packet relay apparatus used to limit the amount of packets transmitted at the transmission source. The packet relay device according to any one of claims 1 to 8 ,
The packet relay device, wherein a total capacity of the buffer memory is smaller than a total queue length of the plurality of queues. A congestion control method in which a packet relay device performs congestion control,
The packet relay device is:
Buffer memory containing multiple queues where packets are temporarily stored;
A storage unit for storing a plurality of first queue lengths indicating the queue lengths of the respective queues, and a second queue length indicating a value corresponding to a total of the plurality of first queue lengths; With
Receiving a packet;
Determining a queue to which the packet belongs from the plurality of queues;
Temporarily storing the packet in the determined queue in the buffer memory and updating the first queue length and the second queue length;
Determining whether to send a congestion notification to the transmission source of the packet according to the first queue length corresponding to the determined queue and the second queue length;
Transmitting the congestion notification to the source;
Equipped with a,
In the step of determining whether to perform the congestion notification, the received packets are classified according to the first queue length, and a threshold value set in advance for each class, the second queue length, Making the determination based on
Congestion control method.

Images